In systems that distribute workloads among multiple compute devices (e.g., in a data center), a centralized server may manage resources allocated to each compute device to efficiently process the workload. In allocating a resource for use by a compute device, the allocation process can consume a significant amount of time, as one or more hardware devices may need to be reinitialized (e.g., reset) to accommodate the remote compute device that is to access the resource. The time consumed in allocating a resource to a compute device in a data center may add latency to the execution of the corresponding workload and potentially result in a violation of a service level agreement (e.g., a set of target performance metrics, such as a speed at which the workload is to be executed, etc.) with a customer associated with the workload.